Research On Detection Method Of Pigment Regions In Face Image

Face skin structure is complex,and the morphology of the pigment regions is different.People judge the subjectivity of pigment regions.The traditional method for detecting pigment regions was to detect a single pigment region collected by dermoscope,which was vulnerable to complex background interference.The detection results were inaccurate and inefficient,such as color clustering detection method and region growth method.The skin detector was used to collect the whole face images,this thesis studied the detection method of multiple pigment regions in the whole face images.The main work of this thesis can be summarized as follows:1.This thesis acquired the method of acquiring image data set.In addition to pigment regions such as nevi and spots,acne was an important factor in judging the health of face skin and interfering with the detection and recognition of pigment regions.Therefore,the main objects of pigment regions detection in this thesis were nevus,spot and acne.Through investigation and research,acne,spot and nevus were mainly manifested in five parts: forehead,left cheek,right cheek,ala nasi and chin.After using the face image collected by the skin detector,this thesis used Landmark to automatically segment the above five main face parts,which formed the local face image data set.2.This thesis presented a method of enhancing face skin image.In order to reduce the influence of local highlights on the detection process,this thesis used a single-scale Retinex algorithm to enhance the local face image.Then,the single-scale Retinex algorithm was compared with the linear gray level change,histogram equalization and adaptive contrast enhancement.The experiment analyzed the contrast between the pigment regions and its normal skin,and the gray mean and gray variance of the whole face image were calculated experimentally.It was found that the single-scale Retinex algorithm could better reduce the local highlights of the image and enhance the boundary of the pigment regions,which was beneficial to the detection of pigment regions.3.This thesis proposed a method for detecting the target pigment regions.A LBF segmentation model with image gradient information was proposed to detect the pigment regions,which integrated local information of image and global information of image gradient,constructed an energy function containing image gradient information.The proposal maked the level set function converge to diversified pigment region boundaries in the evolutionary process.Meanwhile,an edge stop function was added to the perimeter term to restrict the perimeter of the contour that makes the boundary of the pigment region converged more accurately.The experiment results showed that the method proposed in this thesis could greatly improve the accuracy of pigment region detection compared with the methods of maximum class difference,color segmentation,region growth and LBF model,this thesis proposed the method for detecting multiple pigment regions with different color shade and various shapes simultaneously.